Today in AI

Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, released its first in-house model, Inkling — a mixture-of-experts system with 975 billion total parameters that activates only about 41 billion for any given task. Unlike the flagship models from OpenAI, Anthropic, or Google, it's released open-weight, so outside developers can download and modify it directly. The choice reinforces the company's stated bet against "one-size-fits-all" frontier models in favor of something the community can adapt.

OpenAI released a $230 light-up keyboard, co-designed with specialty keyboard maker Work Louder, built specifically to pair with its Codex coding assistant. The launch lands awkwardly alongside Apple's new trade secrets suit over OpenAI's broader hardware ambitions. Niche hardware for a coding agent is an unusual bet, but it fits OpenAI's pattern this year of pushing further into physical products.

Oak came out of stealth with $60 million in funding to tackle a problem more companies are running into: AI agents that need to authenticate as a person, a service, or something in between, without a clean identity system to handle it. The company is building infrastructure to manage which agents can act on whose behalf and with what permissions. As agents get more autonomy, this kind of plumbing is quietly becoming one of the more investable layers of the stack.

Emergent, an AI coding startup based in India, raised a $130 million Series C at a $1.5 billion post-money valuation — just over a year after launching. The speed of the raise reflects how competitive the coding-agent category has become globally, not just among US and Chinese labs. It's also a notable data point for AI startups building outside the usual Bay Area and Beijing axis.